Haruna: A Temporal–Intent Cognitive Scaffold for Stabilizing Multi-Step Reasoning in Large Language Models

This paper proposes Haruna, a symbolic meta-cognitive scaffold that brings temporal rhythm, intentional state representation, and directional meaning into large language model reasoning. Haruna is structured around seven primitives organized into three layers: temporal (hinu–hari), intentional (ral, suha, mināra), and directional-evaluative (vara, luma).Through controlled prompt experiments, ambiguous interpretation tasks, introspective reasoning traces, and educational case studies, Haruna reduces conceptual drift, enhances chronological consistency, and improves clarity when separating facts from interpretations.Haruna is not a claim of artificial consciousness; rather, it serves as a lightweight reasoning interface that can be implemented through prompting alone. The framework aims to contribute to future research on structured cognitive architectures, alignment, interpretability, and human–AI reasoning coherence.